Online Network Detection Of Fertility Of Group Hatched Eggs Based On Machine Vision

In the poultry industry,the hatching of chicken embryo eggs is a time consuming and energy intensive process which takes about 21 days.If the infertile and dead embryo egg is not removed in time,mold within embryo of egg will propagates rapidly that will cause infection of eggs in the whole incubator.Thus,detecting infertile eggs and dead embryo as early as possible can save hatching space and energy,reduce economic losses,and ensure the normal hatching eggs.In this paper,Jingfen No.1 eggs were chosen as experiment material.Online detection platform for fertility of group hatching eggs was established based on machine vision technology.The software was developed to realize the nondestructive detection of fertility of group hatching eggs.The software consists of client and server.The group egg image was sent by the client to the server based on network and was segmented to individual egg image.The image feature of egg was extracted.Then,acquiring fertility of egg by model discrimination.The main research contents and conclusions of this topic are as follows:(1)The construction of platform for nondestructive detection of fertility of group hatching eggs.The PLC as the control module of the detection platform and photoelectric sensor as trigger element were used to realize automatic candling,image collection and egg sorting.The hardware models of camera,lens,light source and PLC were determined.(2)The segmentation of group egg image.Due to dense group egg placement in industrial egg trays,group egg image segmentation is difficult.To solve this,an adaptive group egg segmentation was proposed.The binary image was obtained by Canny algorithm with dynamic threshold and was processed to reduce redundant information.Then,egg shape was obtain by oval fit to segment group egg image.(3)The analysis of egg image feature.The color and texture features of egg image,positions for peak values and peak values of R,G,H,S and I in histogram,peak values of B within histogram,averages,variances and skewness of R,B,G,H,S and I,contrast,correlation,Angular Second Moment(ASM),inverse different moment(IDM),roughness,directionality,line-likeness,regularity,totally 37 feature,were extracted.The T-test was carried out for feature between different kinds of eggs,and the feature with significant difference was taken as the feature variable of the model.(4)The establishment of the model.The Fisher linear discriminant model was established.In order to consider both number and distribution of samples,10 thresholds of Fisher linear discriminant model were proposed.The optimal threshold was selected by comparison of cross validation.Fisher linear discriminant model reached 98.44% and 100% accuracy for 3rd and 4th day hatchling infertile egg,and reached 100% accuracy for tenth day dead embryo egg.The least squares support vector machine was established,which reached 95.31% and 100% accuracy for 3rd and 4th day hatchling infertile egg,and reached 100% accuracy for tenth day dead embryo egg.In conclusion,Fisher linear discriminant model was chosen for detection of infertile and dead-embryo eggs.(5)The development of egg detection software.The software,which realizes the nondestructive detection of fertility of group hatching eggs,was developed by Microsoft Visual Studio and consists of client and server.The client includes the group egg image acquisition module,data transmission module and human-computer interaction module;the server includes egg image processing module,discrimination of fertility of egg module and data transmission module.The Socket was used to complete the network online data transmission,and MSComm control was used to complete the data transmission between computers and PLC.